According to media reports on March 19th, Nvidia CEO Huang Renxun stated that Nvidia's latest AI chip Blackwell will be priced between $30000 (approximately RMB 220000) and $40000. He estimated that Nvidia spent approximately $10 billion (approximately RMB 72 billion) on research and development costs.
CNBC
This price indicates that the chip is likely to be a popular chip for training and deploying artificial intelligence software such as ChatGPT, and its pricing range will be similar to its predecessor H100 (i.e. Hopper).
Huang Renxun v. Kristina Partsinevilos of CNBC, the cost is not only related to chips, but also to designing data centers and integrating them into data centers of other companies.
At 4am Beijing time on March 19th, NVIDIA, the world's third-largest company by market value, officially opened its GTC conference in San Jose, California, USA. Huang Renxun, with his classic leather jacket design, officially unveiled the "nuclear grade" NVIDIA Blackwell architecture and its first chip B200. This dual chip design has 208 billion transistors.
The chip is expected to be shipped later this year.
The Blackwell platform is capable of building and running real-time generative AI on large language models (LLMs) with trillions of parameters, at a cost and energy consumption 25 times lower than its predecessor.
Video screenshot
It is reported that Nvidia claims that Blackwell has six revolutionary technologies that can support models with up to 10 trillion parameters for AI training and real-time LLM inference:
The world's most powerful chip: The Blackwell architecture GPU consists of 208 billion transistors and is manufactured using a customized TSMC 4-nanometer process. Two retail limit GPU bare chips connect 10 TB/s of chip to chip links into a single unified GPU.
Second generation Transformer engine: Combining Blackwell Tensor Core technology with TensorRT LM and Nvidia's advanced dynamic range management algorithm in the NeMo Megatron framework, Blackwell will support double the computational and model size inference capabilities through a new 4-bit floating-point AI.
Fifth generation NVLink: To improve the performance of trillions of parameters and hybrid expert AI models, the latest generation NVLink from Nvidia provides a breakthrough 1.8TB/s bidirectional throughput for each GPU, ensuring seamless high-speed communication between up to 576 GPUs for the most complex LLMs.
RAS engine: Blackwell supports GPUs that include a dedicated engine to achieve reliability, availability, and service. In addition, the Blackwell architecture has added chip level functionality, utilizing AI based preventive maintenance for diagnosing and predicting reliability issues. This can maximize system uptime and improve the resilience of large-scale AI deployments, enabling them to run continuously for weeks or even months, while reducing operational costs.
Secure Artificial Intelligence: Advanced confidential computing capabilities can protect AI models and customer data without compromising performance, and support new native interface encryption protocols, which are crucial for privacy sensitive industries such as healthcare and financial services.
Decompression Engine: A dedicated decompression engine that supports the latest formats, speeds up database queries, and provides the highest performance in data analysis and data science. In the coming years, GPUs will increasingly accelerate the data processing that businesses spend billions of dollars annually.
Video screenshot
The Blackwell GPU has a large volume, combining two individually manufactured grains into a chip manufactured by TSMC. It will also be available as a complete server called GB200 NVLink 2, which combines 72 Blackwell GPUs and other Nvidia components aimed at training AI models.
Amazon, Google, Microsoft, and Oracle will provide access to GB200 through cloud services. GB200 pairs two B200 Blackwell GPUs with an Arm based Grace CPU. Nvidia stated that Amazon Web Services will build a server cluster containing 20000 GB200 chips.
During the Nvidia 2024 GTC, a developer event in the field of artificial intelligence (AI), Nvidia founder and CEO Huang Renxun confidently stated that Nvidia will seize more "cake" in the data center market, which can "dig" billions of dollars a year.
Huang Renxun told the attendees that due to the wide variety of chips and software produced by Nvidia, the company is in a favorable position, and a large part of the global expenditure on data center equipment will come from Nvidia's products.
Huang Renxun predicts that the annual global investment in data center equipment will total $250 billion, with Nvidia products accounting for a larger share than other chip manufacturers. Nvidia is committed to developing software for various industries to adopt and utilize AI technology. Nvidia provides AI models and other software, and then charges customers based on their computing power and the number of chips they run.
As of the close on March 19th local time, Nvidia was at $893.98, up 1.07%, with a market value of $2.2 trillion.
Disclaimer: The content and data in this article are for reference only and do not constitute investment advice. Please verify before use. Based on this operation, the risk is borne by oneself.
